google-deepmind / google-deepmind/formal-conjectures

Proposed proof of source-corrected WOWII Conjecture 160

Open
#4,569 3 comments 0 reactions 0 assignees View on GitHub
Dominant language
Lean
Stars
1.3k
Forks
485
Avg merge
1d 20h
Merged PRs (30d)
328

Description

## Summary

I am submitting an AI-assisted candidate proof of the **source-corrected**
Written on the Wall II Graph Conjecture 160 for independent mathematical
review and possible Lean formalization.

The source correction is tracked in #4423 and PR #4443. In that corrected
statement, $\chi_{C_4}(G)$ is the binary $C_4$-free characteristic: it is
$1$ when $G$ contains no 4-cycle as a subgraph and $0$ otherwise. The
cycle need not be induced.

The candidate theorem is

$$
L_s(G)\ge \max_v\lambda(v)+\max_vT(v)\,\chi_{C_4}(G),
$$

where $\lambda(v)=\alpha(G[N(v)])$.

## Permanent proof package

The proof and verifier are published at permanent commit
[`42878d6c04ec2115e31c01c8e7433da0c07070b4`](https://github.com/vulnix0x4/wowii-conjecture-160/tree/42878d6c04ec2115e31c01c8e7433da0c07070b4):

- [Proof PDF](https://github.com/vulnix0x4/wowii-conjecture-160/blob/42878d6c04ec2115e31c01c8e7433da0c07070b4/proof/WOWII_Conjecture_160_proof.pdf)
- [Proof in Markdown](https://github.com/vulnix0x4/wowii-conjecture-160/blob/42878d6c04ec2115e31c01c8e7433da0c07070b4/proof/WOWII_Conjecture_160_proof.md)
- [LaTeX source](https://github.com/vulnix0x4/wowii-conjecture-160/blob/42878d6c04ec2115e31c01c8e7433da0c07070b4/proof/WOWII_Conjecture_160_proof.tex)
- [Constructive verifier](https://github.com/vulnix0x4/wowii-conjecture-160/blob/42878d6c04ec2115e31c01c8e7433da0c07070b4/verification/verify_wowii160.py)
- [Recorded verification output](https://github.com/vulnix0x4/wowii-conjecture-160/blob/42878d6c04ec2115e31c01c8e7433da0c07070b4/verification/verification_output.txt)

## Proof outline

The proof is constructive.

1. Any tree subgraph of a connected graph can be extended to a spanning tree
without decreasing its number of leaves.
2. If $G$ is $C_4$-free, every open neighborhood is a matching together
with isolated vertices. Consequently,
$d(v)=\lambda(v)+T(v)$ and $\lambda(v)\ge T(v)$.
3. Choose $x$ maximizing $\lambda$ and $y$ maximizing $T$. Two full
stars joined by a geodesic give the required number of leaves in the
general case. The adjacent cases and one exceptional degree-two triangle
case are handled separately.
4. If $G$ contains a 4-cycle, the characteristic term vanishes and a star
formed from a maximum independent set in $G[N(x)]$ gives the bound.

## Computational audit

The accompanying Python program implements the proof's construction; it does
not optimize $L_s$. I reran it independently with NetworkX 3.6.1. It passed:

- all 995 connected unlabeled graphs of orders 2 through 7 in the NetworkX
graph atlas;
- 4,997 deterministic random connected graphs of orders up to 20.

All construction cases in the proof were exercised. This audit is supporting
evidence, not a substitute for proof review.

## Review requested

Could maintainers or graph theorists review the mathematical argument and
confirm that it matches the intended source-corrected invariant? I would also
appreciate guidance on whether a Lean formalization should be completed before
proposing that Conjecture 160 be marked `research solved`.

I am **not** asking to change the status in this issue, and I am not claiming
that the result has been accepted or that historical priority has been
established.

## AI assistance disclosure

OpenAI models assisted with proof search, case analysis, computational
verification, and preparation of the write-up and submission. The proof
package should receive normal independent mathematical and formal review.

Contributor guide

Open the contributing guide

Research direction

Read proof/WOWII_Conjecture_160.md and proof/WOWII_Conjecture_160.tex first, then run verification/verify_wowii160.py and inspect verification/verification_output.txt. Compare the source-corrected invariant and proof cases against the requested independent review; done means a mathematical review is provided or guidance is given on completing Lean formalization.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
documentation
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Quiet
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.